多尺度层级金字塔网络的无人机入侵检测方法

Unmanned Aerial Vehicle Intrusion Detection Method Basedon Multi scale Hierarchical Pyramid Network

近年来,无人机的快速发展给众多领域带来便利,然而无人机入侵给机场安全带来了巨大的挑战。由于无人机目标小、背景复杂、飞行速度快等特点,现有的主流目标检测方法通常难以准确地识别出入侵的无人机,易产生误检漏检的现象。提出了多尺度层级金字塔网络的无人机入侵检测方法,同时利用特征融合模块赋予特征金字塔不同层级、不同尺度的图像语义信息,并通过网格删除和4 Mosaic数据增强技术,对小样本数据集进行扩充,有效地提高了模型的泛化性能。实验表明,方法较于目前最优的无人机检测方法性能提升了5.5%。

In recent years,the rapid development of drones has brought convenience to many fields,but the intrusion of these drones can bring huge challenges for the security of airports.Due to the small target size,complex background,and fast flight speed of drones,existing mainstream methods are difficult to accurately identify drones,leading to false detection and missed detection.This paper proposes a novel multi scale hierarchical pyramid network for unmanned aerial vehicle intrusion detection.The proposed method utilizes feature fusion modules to assign semantic information of images at different levels and scales to the feature pyramid.Through grid deletion and 4 Mosaic data augmentation technology,the small sample dataset is expanded to improve the generalization performance.The experiment shows that the proposed method is improved by 5.5%better compared to the existing object detection methods.